GEN-001 cond. 1: the release step runs make census (and readme-sync) on release/X.Y.Z, with a receipt #4951

Child of E1 #3998, for GEN-001 #4526.

Why. GEN-001 PR1 makes the README contract count equal contracts/census.json .n_files, the release snapshot. Its one writer is make census on the release train (#3569). But no release since 0.69.3 ran it: census.json was last written by the 0.69.3 merge-back (#4338), so 0.70.0 to 0.70.3 shipped n_files: 1837 over a tree of 1895. With PR1 merged and no writer running, the README count would freeze.

Mechanism (not a gate). The release cut or bump step runs make census on release/X.Y.Z and commits contracts/census.json there, then make readme-sync. The merge-back PR carries both to main. The step writes a receipt line with the old and new n_files and the commit.

Done when:

1. The step is in the scripted release path (cut or bump), not a manual note.
2. One dry run on a scratch release/ branch shows census.json and the README both written, with the receipt.
3. It is confirmed that scripts/batch_fold.sh and scripts/prepare_bump.sh never write census.json outside release/X.Y.Z (check_census_derived.sh is_train).

GEN-001 PR1 is not armed until this has its mechanism (C314 Q3 verdict). Related: #4781 (one writer for the derived contract files).

Owner: aprender-clerk. ETA: 48 h from filing.